Comparison of machine learning algorithms for classification of GPS-TEC during Mw>5 earthquakes

The detection of pre-seismic signals has increasingly become an important research area in recent years. Large earthquakes and geomagnetic activity can cause significant disturbances in the Total Electron Content (TEC) of the ionosphere. This study aims to classify disturbances in the ionosphere related to seismic and geomagnetic activity using TEC data. Data obtained for ten major earthquakes with magnitudes between Mw 5,6 to 9,1 occurring between 1999 and 2020 are analyzed for three days prior to and on the earthquake days, geomagnetically quiet and disturbed days periods. various algorithms: Long Short-Term Memory (LSTM), Modular Neural Network (MNN), Probabilistic Neural Network (PNN), Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), Multilayer Perceptron (MLP), RNN-LSTM, CNN-LSTM and CNN-LSTM-MLP are applied to TEC data estimated from Global Positioning System (GPS) stations. While high Accuracy rates are achieved for large earthquakes, the Accuracy of some algorithms decreases for smaller earthquakes, particularly the PNN algorithm performes poorly for medium-sized earthquakes. The study indicates that ionospheric disturbances associated with large earthquakes can be effectively classified, but current algorithms are insufficient for smaller earthquakes. Future work should focus on developing more advanced algorithms for smaller earthquakes and geomagnetic activity.